* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Seven unique food consumption patterns identified among
Food safety wikipedia , lookup
Overeaters Anonymous wikipedia , lookup
Food and drink prohibitions wikipedia , lookup
Raw feeding wikipedia , lookup
Human nutrition wikipedia , lookup
Food coloring wikipedia , lookup
Obesity and the environment wikipedia , lookup
Vegetarianism wikipedia , lookup
Food studies wikipedia , lookup
European Journal of Clinical Nutrition (2000) 54, 314±320 ß 2000 Macmillan Publishers Ltd All rights reserved 0954±3007/00 $15.00 www.nature.com/ejcn Seven unique food consumption patterns identi®ed among women in the UK Women's Cohort Study DC Greenwood1*, JE Cade1, A Draper2, JH Barrett 3, C Calvert1 and A Greenhalgh1 1 Nuf®eld Institute for Health, 71 ± 75 Clarendon Road, University of Leeds, Leeds LS2 9PL, UK; 2Public Health Nutrition Unit, Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, 49 ± 51 Bedford Square, London WCIB 3DP, UK; and 3ARC Epidemiology Unit, University of Manchester, Manchester M13 9PT, UK Objective: To identify groups of subjects with similar food consumption patterns so that complex disease ± diet relationships can be investigated at the level of the whole diet, rather than just in terms of nutrient intake. Subjects: 33,971 women in the UK Women's Cohort Study. 60,000 women on the World Cancer Research Fund mailing list were initially invited to take part. Subjects were selected to include a high proportion of vegetarians. Design: The cohort completed a 217 item food frequency questionnaire. Cluster analysis was used to identify groups of women with similar food consumption patterns. Clusters were compared on socio-demographic characteristics, indicators of health and diet, and nutrient intakes. Results: Seven clusters were identi®ed including two vegetarian clusters. Groups appeared to be differentiated by differences in food types and in diversity of diet. Socio-demographic, health and diet characteristics and nutrient intakes all differed signi®cantly between groups. Conclusion: Classifying diets in more pragmatic terms than just nutrient intake should provide valuable insight into understanding complex diet-disease relationships. Dietary advice, whilst based on nutrient content of meals, needs to take account of the combinations of different food types that people naturally choose to use together. Sponsorship:World Cancer Research Fund. Descriptors: cohort studies; multivariate analysis; diet; food habits European Journal of Clinical Nutrition (2000) 54, 314±320 Introduction In nutritional research it is more common to consider food consumption in terms of nutrient intake, rather than type of food consumed. Conventionally nutrient intake is compared with recommended levels or adherence to dietary guidelines to identify groups of people with various levels of intake of speci®c nutrients. These groups can then be described in terms of their socio-demographic characteristics. This enables the researcher to identify individuals or more often subgroups of the population whose diets may be nutritionally de®cient in some way. Whilst researchers have concentrated on classifying people on the basis of their nutrient intake, consumers consider more than nutrient content when choosing food. In addition it is unlikely that the aetiology of diseases such as cancer can be explained by levels of single nutrients and it may be that non-nutritive substances, such as phytochemicals, may be involved. Most consumers eat foods in certain combinations or patterns, and interest is therefore now *Correspondence: DC Greenwood, Nuf®eld Institute for Health, 71 ± 75 Clarendon Road, University of Leeds, Leeds LS2 9PL, UK. Contributors: DCG wrote the paper, performed statistical analyses, and interpreted results. JEC initiated the study, designed the study, assisted in interpreting results and assisted in writing the paper. AD sits on the steering group, advised on analysis, assisted in interpreting results, and assisted in writing the paper. JHB sits on the steering group, performed early statistical analyses, assisted in interpreting results and writing the paper. CC assisted in data collection, maintained the cohort, assisted in statistical analyses and editing the paper. AG maintained the cohort, assisted in statistical analyses, and asisted in interpreting results and editing the paper. Received 29 July 1999; revised 19 October 1999; accepted 20 October 1999 focussing on whole diets and lifestyles rather than single nutrients. Identi®cation of these patterns would be useful to explore complex diet-disease relationships (such as investigating potential effects of the Mediterranean diet) and also to intervene to provide relevant nutritional advice and education. Cluster analysis has been used in similar contexts to identify groups or clusters of people with similar characteristics (Wirfalt & Jeffery, 1997; Schroll et al, 1996; Tucker et al, 1992; Hulshof et al, 1992; Bisgrove et al, 1989; Farchi et al, 1989; Akin et al, 1986; Windham et al, 1985). We aimed to classify people in terms of their chosen patterns of food consumption, rather than nutrient intake, to identify broad food groupings that naturally tend to be used together. This would also identify food types that are consumed to the exclusion of others. In other words, we hoped to identify consumption patterns or habits that classify people according to the types of food they choose to consume, forming natural groupings of people characterized by similar food consumption patterns. In this investigation we used the UK Women's Cohort Study, a large national cohort funded by the World Cancer Research Fund (WCRF), to investigate food consumption patterns. Methods The UK Women's Cohort Study consists of 33,971 women aged 35 ± 69y at recruitment, identi®ed from responses to a brief questionnaire sent to potential supporters by the WCRF. Between 1995 and 1998, 61,000 potential subjects received a 217 item food frequency questionnaire (FFQ), Seven unique food consumption patterns DC Greenwood et al which had been adapted from the FFQ used in the European Prospective Investigation into Cancer (EPIC) study (Riboli, 1992). Subjects were selected to ensure a high proportion of vegetarians in the cohort. All self-de®ned vegetarians were chosen and matched for age to the next nearest red-meat eater. All meat eaters not eating red meat were also included. In this way we ensure a range in intake of foods and nutrients, with greater numbers of people consuming larger quantities of vegetables and fruit, and so be more likely to detect statistically signi®cant associations between diet and disease outcomes. In particular the design maximizes power to investigate associations between vegetarian or non-red meat diets and disease outcomes. Frequency of consumption of each food item in the previous 12 months was recorded on the same 10 point scale: `never', `less than once a month',`1 ± 3 per month', `once a week', `2 ± 4 per week', `5 ± 6 per week', `once per day', `2 ± 3 per day', `4 ± 5 per day', and `6+ per day'. For the purposes of this investigation, food items similar in type and nutrient content were merged to reduce the number of food items from 217 to 74 food types to ease interpretation and computation of the clusters. The number of clusters into which the women's food use was grouped was identi®ed using a scree plot. This showed for different numbers of clusters how well the groups represented the individuals assigned to them in terms of the sum of squared Euclidean distances between individuals and their particular cluster mean (measured by the trace of the within-groups dispersion matrix, Everitt, 1993). Increasing the number of clusters decreases the trace, showing improved ®t. The number of clusters that the women were naturally grouped into was indicated by the number of clusters that gave the biggest improvement in ®t compared to the previous clusters. This provided an objective method of determining the number of clusters. Cluster analysis was performed using SPSS version 8. The k-means cluster algorithm (Everitt, 1993; Anderberg, 1973) was used because of the large size of the sample and to avoid imposition of a hierarchical structure on the clusters. It assigns each woman automatically to an initial cluster characterising a particular food consumption pattern closest to their own and then updates the cluster characteristics to represent all the women assigned to it, including the new individual. After updating all the clusters for all the women, the process then repeats, reassigning each woman to the nearest updated cluster. This continues until the characteristics of the clusters stabilize and the women in them remain unchanged. Through this iterative process the women will have been assigned to the cluster that most characterizes their patterns of food consumption. Robustness of the results was assessed by trying different initial cluster centres, and different rules of convergence for the iterations. The variables were not standardized to unit variance prior to the cluster analysis because all the variables were measuring the same thing (consumption), standardizing would dilute differences between clusters, and standardizing would also ignore correlations between variables (Everitt, 1993). The sensitivity of the results to the method used was assessed by repeating the analysis with standardized variables. The food consumption patterns, socio-demographic characteristics, indicators of health and diet, and nutrient intakes of the different groups of women identi®ed by the cluster analysis were tabulated. The characteristics of the clusters which were measured by continuous variables were compared using analysis of variance (and non-parametric Kruskal ± Wallis analysis of variance where appropriate). Characteristics measured by dichotomous variables were compared using chi-squared tests. Ethics approval for the UK Women's Cohort Study was obtained individually from all 174 Local Research Ethics Committees representing all the women in the cohort. 315 Results Description of cohort The mean age of the women in the cohort was 52 y (standard deviation (s.d.) 9 y). The women were generally well educated, with 14,749 (43%) being educated to `A' level standard or beyond. The mean age of leaving full-time education was 18 y (s.d. 4 y). A total of 18,217 (55%) of the women were in a managerial, administrative, professional, technical or associated professional socio-economic group. Only 3608 (12%) reported levels of fruit and vegetable consumption meeting or exceeding the UK Department of Health recommendation of ®ve portions of fruit and vegetables per day (Department of Health, 1994). The women's exercise levels, however, were high. The median length of time spent doing physical activity vigorously enough to cause sweating or raise the heartbeat was 60 (inter-quartile range (IQR); 0 ± 120) min per week. These ®gures are supported by slightly lower body mass indexes (BMIs) for our women, with a mean of 24.4 kg=m2 (s.d. 4.3 kg=m2) compared to those in the National Diet and Nutrition Survey (Gregory et al, 1990), 24.8 kg=m2 for women aged 35 ± 49 and 26.2 kg=m2 for women aged 50 ± 64). Based on a de®nition of obesity being BMI 30 kg=m2, just 3095 (9%) were obese. The mean waist-to-hip ratio was also low at 0.75 (s.d. 0.06). A total of 9793 (30%) out of 32,849 who answered the question described themselves as vegetarian. A high proportion of the cohort (17,920 women, 58%) take vitamins, minerals, ®sh oils, ®bre or other food supplements. This is a substantially higher proportion than in the National Diet and Nutrition Survey (Gregory et al, 1990). Only 3670 (11%) reported they were current smokers, either regular or occasional. Cluster analysis Increasing the number of clusters to seven produced the largest improvement in the trace, suggesting that the women were naturally grouped into seven clusters. The cluster centres were examined and the groups subjectively given the labels `monotonous low-quantity omnivores', `health conscious', 'traditional meat chips and pudding eaters', `higher-diversity, traditional omnivores', `conservative omnivores', `low diversity vegetarians' and `high diversity vegetarians'. The food consumption characteristics of these groups are shown in Table 1. The food groups identi®ed as having relatively high or low levels of consumption for a particular cluster, marked out that cluster from the others. Certain food items from the 217 item FFQ illustrate these differences well. Typical food consumption frequencies for a selection of food items are shown for the seven clusters in Table 2. The food items have been selected for illustrative purposes, out of those that differentiate most clearly between the clusters, whilst covering a range of food types. European Journal of Clinical Nutrition Seven unique food consumption patterns DC Greenwood et al 316 Table l Food consumption characteristics of the seven clusters Label Number of women Characteristic food types consumed High quantities Moderate quantities 1. Monotonous low-quantity omnivores 5416 White bread, milk, sugar 2. Health conscious 2131 3. Traditional meat, chips and pudding eaters 6087 4. Higher diversity, traditional omnivores 4819 5. Conservative omnivores 5946 6. Low diversity vegetarians 5190 7. High diversity vegetarians 4379 Bran, potatoes, wholemeal food, Most other foods including yoghurt, low-fat dairy products, meat and wine pulses, ®sh, vegetables, salad, fruit White bread, chips, meat, sugar, Most other foods including Wholemeal food, soya products, high-fat and creamy food, alcohol vegetables, salad, fruit biscuits, cakes Chips, white pasta and rice, Vegetbles, fruit, alcohol Less cakes and puddings than 3 high-fat and creamy food, eggs, meat, ®sh, chocolate, biscuits, crisps. More ®sh and salad than 3. More diversity than 3 Most food, including potatoes, Cereals, chips, wholemeal food, nuts, meat, ®sh, eggs, fruit, vegetables pulses, spreads and dressings, chocolate, crisps, biscuits. Less red meat, less chips and less puddings than 3 and 4 Wholemeal bread, soya products, Cereals Butter, eggs, meat, ®sh pulses, fruit (apart from exotic fruit), vegetables Wholemeal bread, cereals, White bread, meat, ®sh wholemeal pasta and rice, soya products, spreads, nuts, pulses, vegetables, fruit, herbal tea (generally higher consumption of these products than 6) The characteristics of the clusters were essentially unchanged by using different initial cluster centres or convergence rules. In addition, clusters with similar interpretations were obtained using the less sensitive but more robust approach of standardizing the variances of the variables before performing the cluster analysis, demonstrating the robustness of the results. The clusters whose characteristics were most similar (ie whose centres were closest together) were cluster 3 (traditional meat, chips and pudding eaters) and cluster 5 (conservative omnivores), though cluster 4 (higher-diversity, traditional omnivores) was also relatively close to the other two. The clusters whose centres were furthest apart were cluster 1 (monotonous low-quantity omnivores) and cluster 2 (health conscious). Comparison of clusters Comparison of the socio-demographic characteristics of the groups identi®ed by the clusters is shown in Table 3. Vegetarian and health conscious groups were generally better educated, smoked less, and contained more women in higher socio-economic groups. Indicators of reported vegetarian status, health and healthy lifestyle are shown in Table 4. Women with healthier food consumption patterns had lower BMIs and, in particular, both the vegetarian clusters had the lowest mean BMIs. However, the mean BMIs for all groups were quite low, and levels of obesity were around 10% for all but the vegetarian groups, which were 5 ± 6%. The proportion of women meeting the UK Department of Health's guidelines for eating ®ve portions of fruit and vegetables per day (Department of Health, 1994) was assessed using a cross-check question to avoid over-reporting of fruit European Journal of Clinical Nutrition Potatoes, meat Low quantities Most other foods, especially bran and wholemeal food, soya products, yoghurt, nuts, pulses, fruit and vegetables, alcohol Chips, sugar and vegetable consumption in the FFQ (Calvert et al, 1997), although this may have led to some under-reporting. Group 1, labelled `monotonous low-quantity omnivores' were least likely to meet the guidelines for eating ®ve portions of fruit and vegetables per day. Women with healthier food consumption patterns were also more likely to take vigorous physical exercise (activity vigorous enough to cause sweating or raise the heartbeat). Women with less healthy eating patterns such as the `traditional meat, chips and puddings' group were the least likely to take vitamins and other food supplements, although overall levels were still high. The two consumption patterns identi®ed in the cluster analysis which are labelled as vegetarian patterns contain some women who do not describe themselves as vegetarians. However, these women are less likely to eat red meat than non-vegetarians in the other clusters (71% vs 98% eating red meat, P < 0.00l), and where they do eat red meat, they eat substantially less (mean of 0.28 portions per day vs 0.87, P < 0.00l). The ®ve clusters not labelled as vegetarian do contain some women who call themselves vegetarians. Of these people who say they are vegetarians, 59% do eat meat, most of which is ®sh. Only 12% a of the women in these clusters who call themselves vegetarians eat red meat, and those women who do eat red meat do so less than those in the same clusters who describe themselves as nonvegetarians (0.02 portions per day vs 0.87 portions per day, P < 0.001). Daily nutrient intakes for each cluster are shown in Table 5 for a range of important nutrients. The `health conscious' group had the lowest percentage of total energy intake from fat (30%), and the `traditional meat, chips and pudding eaters' had the highest (35%) despite similar Less a Less a Less a Once per week Once per week 2 ± 4 per week 2 ± 4 per week Once per week Never 1 ± 3 per month Once per week Once per week 1 ± 3 per month 1 ± 3 per month Less a Less a Less a Once per week 2 ± 4 per week 2 ± 4 per week Once per day 3. Traditional meat, chips and pudding eaters 4. Higher diversity traditional omnivores 5. Conservative omnivores 6. Low diversity vegetarians 7. High diversity vegetarians Once per day than once month than once month than once month Less than once a month Once per week Once per day Less than once a month Less than once a month 1 ± 3 per month Less than once a month Never 1 ± 3 per week Once per week 2 ± 4 per week than once month than once month than once month 1 ± 3 per month 2 ± 4 per week 2 ± 4 per week Once per week Once per week 1 ± 3 per month 1 ± 3 per month Once per week 1 ± 3 per month 1 ± 3 per month 1 ± 3 per month Once per week 1. Monotonous low quantity omnivores 2. Health conscious Less a Less a Less a than once month than once month than once month 1 ± 3 per month 5 ± 6 per week 5 ± 6 per week Less than once a month Less than once a month 1 ± 3 per month Less than once a month Less than once a month 2 ± 4 per week Less than once a month Once per week Less than once a month Never Never Oranges Peppers Celery Roast beef Chips Wholemeal bread Label Pecan nuts and walnuts 1 ± 3 per month Sponge cake Plums Puddings Fruit Vegetables Meat Nuts Potatoes Bread Median food frequencies Food type: Table 2 Food frequencies for an illustrative selection of food items from the 217 item food frequency questionnaire. Food items have been selected for illustrative purposes, out of those that differentiate most clearly between the clusters, whilst covering a range of food types. The cluster(s) with the most frequent consumption of each item is(are) shown in bold type Seven unique food consumption patterns DC Greenwood et al absolute values. This was due to higher carbohydrate intakes diluting the effect of fat in the `health conscious' group. The `conservative omnivores' also had 30% of total calori®c intake from fat with a low total fat intake. There were statistically signi®cant differences (P < 0.001) between the seven clusters for all the characteristics listed in Tables 3 ± 5, even where the sizes of the differences were of little practical importance. This is because of the large sample size. Attention should be focussed on the estimates, rather than the P-values. 317 Discussion The women in general had healthier diets, healthier lifestyles and were better educated than would be seen in a random sample of UK women of the same age. This is largely because of the source of the cohort from the WCRF questionnaire data targeted at potential supporters, and to include a large proportion (one-third) of vegetarians. It is also to be expected that women most interested in taking part in a study such as this would be those who have a keener interest in diet and health. As such, the cohort forms a unique group of women, and this context needs to be taken into account when considering the clusters identi®ed. The food consumption patterns, however, are likely to be relevant to all UK women because they are drawn from internal comparisons, although the relative size of each cluster may not be applicable outside this cohort. While previous studies have identi®ed diets with similar nutritional content, this analysis has identi®ed patterns of food choice. The FFQ on which the patterns are based covers a 12 month period, but it is known that individuals' dietary behaviours and patterns do change over longer periods of time (Ministry of Agriculture, Fisheries and Food, 1991). It may therefore be expected that the basic groupings identi®ed by the clusters will remain, but that over time individuals may change which group most closely re¯ects their eating habits. Repeated food frequency questionnaires to the cohort over 10 y will facilitate exploration of individuals' patterns and habits which are more resistant to change than others. This may provide a focus for targeting nutritional advice. Interpretation and labelling of clusters has a large subjective component. In addition, the inclusion of nonvegetarians in the two vegetarian groups and vegetarians in the clusters labelled as non-vegetarian could be seen as a limitation of the method. However, investigating these apparent inconsistencies further reveals that the women's food consumption patterns of some vegetarian women are closer overall to some non-vegetarians. The results suggest that for many their diet may be similar to non-vegetarians, without the diversi®cation into other food types more common amongst vegetarian diets, but with red meat consumption reduced or cutout, prompting them to call themselves vegetarians. Similarly, the non-vegetarians who have been placed in one of the two vegetarian groups have consumption patterns which overall are more similar to the vegetarian groups than the non-vegetarian. This could indicate a group who follow a vegetarian diet as far as possible, but occasionally allow themselves ®sh. The FFQ is based on frequency and range of foods consumed. The key differences between the patterns of food choice identi®ed by the cluster analysis are based not just on differences in food types, but also in diversity of diet. For example, two separate groups of vegetarians were European Journal of Clinical Nutrition Seven unique food consumption patterns DC Greenwood et al 318 Table 3 Socio-demographic characteristics of the seven clusters Meam age of leaving `A' level Managerial administrative, full-time education education professional or technical Mean age (s.d.) Current Smoker (%) (s.d.) or higher (%) job (%) Label 1. Monotonous low quantity omnivores 2. Health conscious 3. Traditional meat, chips and pudding eaters 4. Higher-diversity, traditional omnivores 5. Conservative omnivores 6. Low diversity vegetarians 7. High diversity vegetarians 53 53 52 53 55 49 50 (10) (9) (9) (9) (9) (9) (9) 903 163 754 393 548 574 333 (17%) (8%) (13%) (8%) (9%) (11%) (8%) 17 19 18 18 18 19 19 (3) (4) (3) (3) (3) (4) (4) 1594 1015 2176 2181 2300 2847 2636 (29%) (48%) (36%) (45%) (39%) (55%) (60%) 2319 1344 2782 2621 3088 3139 2924 (44%) (64%) (46%) (55%) (53%) (61%) (68%) Percentages for each question are based on the number of women completing that question. Table 4 Additional health and dietary characteristics of the seven clusters Label 1. Monotonous low quantity omnivores 2. Health conscious 3. Traditional meat, chips and pudding eaters 4. Higher diversity, traditional omnivores 5. Conservative omnivores 6. Low diversity vegetarians 7. High diversity vegetarians Describes Obese (%) Use food herself as Mean BMI (BMI 30 Mean waist-to- supplements vegetarian (%) (s.d.) kg/m2) hip ratio (s.d.) (%) Consume recommended Median time 5 portions of fruit in vigorous physical and vegetables activity (minutes per day (%) per week (IQR)) 1057 (20%) 25 (5) 602 (12%) 0.75 (0.07) 2621 (53%) 197 (5%) 0 (0 ± 120) 709 (35%) 198 (3%) 24 (4) 25 (4) 192 (9%) 724 (12%) 0.74 (0.08) 0.75 (0.06) 1243 (65%) 2809 (50%) 551 (30%) 232 (4%) 70 (0 ± 180) 20 (0 ± 120) 226 (5%) 25 (4) 465 (10%) 0.75 (0.06) 2437 (55%) 517 (12%) 60 (0 ± 120) 940 (16%) 3517 (71%) 3146 (75%) 25 (4) 23 (4) 23 (4) 637 (11%) 277 (6%) 198 (5%) 0.75 (0.06) 0.74 (0.06) 0.74 (0.06) 3004 (56%) 3102 (66%) 2704 (69%) 715 (14%) 531 (12%) 865 (23%) 30 (0 ± 120) 60 (60 ± 120) 60 (0 ± 180) Percentages for each question are based on the number of women completing that question. identi®ed. The type of food they consumed was similar in that they both chose vegetarian diets, but one group consumed a wider variety of such items than the other. Whilst both vegetarian clusters and the women with health conscious behaviour apparently more easily met recommended levels of fruit and vegetable consumption, we have also identi®ed groups of women with eating patterns for which these levels are harder to attain. In particular, in addition to the type of food chosen, it appears that the level of diversity is important here. Encouraging movement from a monotonous, conservative diet to a more diverse one could be one way to help people improve the quality of their diet, within the framework of their personal food preferences or restrictions. The poorer food consumption patterns were also re¯ected in slightly higher levels of obesity, although a causal relationship cannot necessarily be implied. Previous work identifying dietary patterns in two US Metropolitan populations found that food choice patterns with more evenly distributed food energy sources were related to lower BMIs and more favourable nutrient intakes (Wirfalt & Jeffery, 1997). In our study, the mean BMI levels differed by only about 2 kg=m2, with the vegetarian groups being slightly lower than the other groups. In our cohort the groups displaying greater diversity of food choice also have fewer obese women and more favourable nutrient intakes than those labelled `monotonous', `conservative' or `low diversity', in broad agreement with this previous work. Our analysis therefore adds to a growing interest in the effects of and the concepts surrounding dietary diversity (Dowler & Calvert 1995; Krebs-Smith et al, 1987). It is possible that diversity is in¯uenced by cost (Dowler & Calvert 1995; Cade 1999; Ministry of Agriculture, Fisheries and Food, 1996; Bolton-Smith et al, 1991). European Journal of Clinical Nutrition The health conscious and vegetarian groups were more likely to have a higher educational level and do more vigorous exercise than the other groups. They were also more likely to take food supplements consistent with the hypothesis that supplement use is associated with a healthier lifestyle pro®le and an adequate nutritional intake (Kirk et al, 1999; Draper et al, 1993). Nutrient intakes were overall quite high. The `health conscious' group had the highest intakes of the vitamins and minerals considered. Their energy intakes were similarly high to the `higher diversity traditional omnivores' who were also high quantity consumers, but the health conscious group had a lower percentage of their energy coming from fat (30% vs 34% respectively). The lowest nutrient intakes were seen in the `monotonous low-quantity omnivores' group. Dietary patterns have been described for people living in the Mediterranean region. Trichopoulou et al (1995) created a diet score for elderly subjects in rural Greece based on eight components characteristic of the Mediterranean diet (Trichopoulou et al, 1995). A higher score indicated a more typically Mediterranean diet. They found that a one unit increase in diet score was associated with a signi®cant (17%) reduction in overall mortality. The individual components of the diet score had weak and generally nonsigni®cant associations with survival, implying that it is the pattern of consumption which maybe more important in health terms. We intend investigating associations between the maintenance of consumption patterns identi®ed in our study and subsequent incidence of cancer over a 10 y period. Not all diets which are potentially healthy in one aspect, eg low-fat, will be healthy in others, eg micronutrient intake. It may be that high-fat diets which are high in 87 83 96 10 (3) 10 (3) 12 (4) 79 73 92 17 (7) 17 (7) 22 (8) 89 89 98 1123 (538) 1029 (421) 1293 (533) 100 100 100 51 51 52 30 32 33 1842 (576) 1. Monotonous low quantity omnivores 2. Health conscious 3. Traditional meat, chips and pudding eaters 4. Higher-diversity, traditional omnivores 5. Conservative omnivores 6. Low diversity vegetarians 7. High diversity vegetarians 80 (22) 76 (22) 91 (26) 16 14 14 8.5 (2.1) 9.2 (2.4) 11.2 (3.0) 2033 (508) 2199 (585) 2682 (724) 68 (22) 79 (28) 97 (34) 278 (82) 297 (87) 372 (107) 192 (76) 160 (64) 236 (86) 100 14 (4) 95 22 (8) 100 1632 (623) 100 48 34 15 12.4 (3.3) 112 (29) 2978 (789) 113 (36) 383 (111) 226 (78) 97 98 14 (5) 12 (4) 96 80 26 (10) 18 (7) 99 97 1623 (818) 1308 (560) 100 100 54 47 30 35 14 15 2957 (1063) 12.4 (4.4) 106 (38) 2521 (671) 10.5 (2.8) 96 (26) 97 (47) 98 (32) 427 (154) 316 (94) 383 (172) 149 (62) 73 9(3) 60 14 (7) 74 922 (532) 97 113 (61) 49 241 (87) 32 66 (26) 16 72 (22) g (s.d.) MJ (s.d.) kcal (s.d.) Label 7.7 (2.4) mg (s.d.) Percentage Percentage Percentage who meet who meet who meet RNI g (s.d.) RNI g (s.d.) RNI Percentage Percentage energy who meet intake mg (s.d.) RNI Percentage Percentage energy energy g (s.d.) intake g (s.d.) intake Vitamin C Carbohydrate Fat Protein Energy intake (including alcohol) Table 5 Mean (s.d.) total daily intake of energy (including alcohol) and seven clusters with percentage meeting Reference Nutrient Intakes (RNI) Vitamin A Iron Zinc Seven unique food consumption patterns DC Greenwood et al fruit and vegetables are associated with lower health risks than similar fat intakes with low fruit and vegetable intakes (Wirfalt & Jeffery, 1997). There is a danger of focussing on single nutrients when dietary intake is a multidimensional behaviour (Ursin et al, 1993; Patterson et al, 1994). Therefore investigating whole diets as in this study, taking into account not only frequency of eating speci®c items but also the range and diversity, may better describe groups of people in different risk groups. In particular, the identi®cation of two distinct types of vegetarian diet is useful in this regard. Following our approach, classifying diets in more pragmatic terms than just nutrient intake should provide valuable insight into understanding complex diet ± disease relationships. Effects of diet are more than just the nutrients they contain; there are complex interactions between nutrients, other dietary constituents, and the processing and preparation of food. Information about such interactions is largely lost in conventional dietary analyses, which ignore the different patterns of food intake. Associating the maintenance of consumption patterns identi®ed in this study with subsequent cancer incidence will take these interactions into account, allowing early dietary recommendations to be made. This will also provide pointers towards the underlying mechanisms, informing the investigation of the complex underlying interactions. In addition, dietary advice, whilst based on nutrient content of meals, needs to take account of the combinations of different food types that people naturally choose to use together. Identifying these food consumption patterns will inform the development of health interventions aiming to improve people's diets. 319 Acknowledgements ÐThe UK Women's cohort is funded by the World Cancer Research Fund. We thank the members of the steering group: David Forman, Barrie Margetts, Margaret Thorogood and Rhys Williams. We also thank Cathy Britton, Barbara Bailey and Carole Burton for data processing. References Akin JS, Guilkey DK, Popkin BM & Fanelli MT (1986): Cluster analysis of food consumption patterns of older Americans. J. Am. Diet. Assoc. 86, 6l6 ± 624. Anderberg MR (1973): Cluster Analysis for Applications. New York: Academic Press. Bisgrove EZ, Popkin BM & Barba C (1989): Infant feeding in the Philippines: a cluster analysis approach. Ecol. Food. Nutr. 23, 75 ± 90. Bolton-Smith CSW, Woodward M & Tunstall-Pedoe H (1991): Nutrient intakes in different social class groups: results from the Scottish heart health study. Br. J. Nutr. 65, 321 ± 325. Cade JE, Upmeier H, Calvert C & Greenwood D (1999): Costs of a healthy diet: analysis from the UK Women's Cohort Study. Public Health Nutr. (in press). Calvert C, Cade J, Barrett JH & Woodhouse A (1997): Using cross-check questions to address the problem of mis-reporting of speci®c food groups on food frequency questionnaires. Eur. J. Clin. Nutr. 51, 708 ± 712. Department of Health (1994): Nutritional Aspects of Cardiovascular Disease. Report on Health and Social Subjects No. 46. London: HMSO. Dowler E & Calvert C(1995): Nutrition and diet in lone-parent families in London. London: Family Policy Studies Centre. Draper A, Lewis J, Malhotra N & Wheeler E (1993): The energy and nutrient intakes of different types of vegetarian: a case for supplements? Br. J. Nutr. 69, 3 ± 19. Everitt BS(1993): Cluster Analysis. 3rd ed. London: Edward Arnold. Farchi G, Mariotti S, Menotti A, Seccareccia F, Torsello S & Fidanza F (1989): Diet and 20-year mortality in two rural population groups of middle-aged men in Italy. Am. J. Clin. Nutr. 50, l095 ± 1l03. European Journal of Clinical Nutrition Seven unique food consumption patterns DC Greenwood et al 320 Gregory J, Foster K, Tyler H & Wiseman M (1990): The Dietary and Nutritional Survey of British Adults. London: HMSO. Hulshof KFAM, Wedel M, Lowik MRH, Kok FJ, Kistemaker C, Hermus RJJ, ten Hoor F & Ockhuizen T (1992): Clustering of dietary variables and other lifestyle factors (Dutch nutritional surveillance system). J. Epidemiol. Community Health 46, 417 ± 424. Kirk SFL, Cade JE, Barrett JH & Conner M (1999): Diet and lifestyle characteristics associated with dietary supplement use in women. Public Health Nutr. 2, 69 ± 73. Krebs-Smith SM, Smiciklas-Wright H, Guthrie HA & Krebs-Smith J (1987):The effects of variety in food choices on dietary quality. J. Am. Diet. Assoc. 87, 897 ± 903. Ministry of Agriculture, Fisheries and Food (1991): Fifty years of the National Food Survey, 1940-1990. The Proceedings of a Symposium held in December 1990. London: HMSO. Ministry of Agriculture, Fisheries and Food (1996): National Food Survey. London: HMSO. Patterson RE, Haines PS & Popkin BM (1994): Diet Quality index: Capturing a multidimensional behavior. J. Am. Diet. Assoc. 94, 57 ± 64. Riboli E(1992): Nutrition and cancer: Background and rationale of the European Prospective Investigation into Cancer and Nutrition (EPIC). Ann. Oncol. 3, 783 ± 791. European Journal of Clinical Nutrition Schroll K, Carbajal A, Decarli B, Martins I, Gruneberger F, Blauw YH & de Groot CP (1996): Food patterns of elderly Europeans. Eur. J. Clin. Nutr. 50(Suppl 2), S86 ± 100. Trichopoulou A, Kouris-Blazos A, Wahlqvist M, Gnardellis C, Lagiou P, Polychronopoulos E, Vassilakou T, Lipworth L & Trichopoulos D (1995): Diet and overall survival in elderly people. Br. Med. J. 311, l457 ± 1460. Tucker KL, Dallal GE & Rush D(1992): Dietary patterns of elderly Boston-area residents de®ned by cluster analysis. J. Am. Diet Assoc. 92, 1487 ± 1491. Ursin C, Ziegler RG, Subar AF, Craubard BI, Haile RW & Hoover R (1993): Dietary patterns associated with a low-fat diet in the National Health Examination Follow-up Study: identi®cation of potential confounders for epidemiologic analyses. Am. J. Epidemiol. 137, 916 ± 927. Windham CT, Windham MP, Wyse BW & Hansen RG (1985): Cluster analysis to improve food classi®cation within commodity groups. J. Am. Diet Assoc. 86, 1306 ± 1314. Wirfalt AK & Jeffery RW (1997): Using cluster analysis to examine dietary patterns: nutrient intakes, gender, and weight status differ across food pattern clusters. J. Am. Diet Assoc. 97, 272 ± 279.